Insurance leaders are drawing a hard line between AI execution and human judgment. A survey of 150 sector decision-makers, commissioned by AI-powered automation provider mea Platform and conducted by Information Services Group, found 83% back AI for repeatable operational tasks, while 86% insist people must retain authority over consequential decisions. The findings point to a growing need for insurers to define precisely where that boundary sits before scaling further - distinguishing tasks like extracting a policy term from decisions like binding a non-standard risk.
The research lands as nearly all insurers, 96%, have some form of AI-led operating model redesign on their agenda. Yet only about 13% currently operate at an advanced AI posture, even though 52% intend to reach that level. "Insurers know exactly where they want the line between AI and human judgment. Getting there is the part they have not solved," said Ashish Jhajharia, insurance SME and principal analyst with ISG. "The gap between 96% with redesign on the agenda and the 13% that have reached an AI-centred operating model defines the next two years."
A four-owner operating model
The report recommends a four-owner structure: AI executes repeatable work, people hold judgment and accountability, global capability centers build enterprise capability, and outside partners deliver contracted services. The institution retains accountability for outcomes across all four lanes. This model reflects a view that scaling AI in insurance requires architectural clarity, not just technology deployment.
Trust in AI systems remains highly conditional. Three-quarters of respondents said they most trust insurance-specific or governed-hybrid AI approaches. Only 6% trust general-purpose models alone. Domain knowledge, the report said, does not substitute for governance.
The report outlined what "governed" should mean in practice: connecting AI workflows to controlled policy wording and underwriting appetite, separating what a system may read from what it may execute, ensuring human reviewers have sufficient time and authority to challenge outputs, and reassessing authority whenever policy wording, data or the underlying model changes. Permission to operate with reduced oversight should follow demonstrated evidence of decision quality and appropriate referrals - not just improvements in speed or cost. If performance drops below agreed thresholds, execution should be restricted.
Ambition outpaces execution
Among insurers using business process outsourcing partners, just 12% reported broad or AI-first service deployment. The majority, 78%, said they are using production AI mainly for incremental efficiency gains or limited automation. This gap between ambition and delivered execution has direct operational consequences. ISG estimated that roughly one in nine broker submissions is declined or left unquoted because of capacity constraints rather than risk appetite.
Insurers reported productivity gains of 61% and faster cycle times of 51% from AI deployment to date. They expect an average cost reduction of 16% over 24 months. Yet the activity-level analysis of 20 insurance operations found the AI-human boundary varies sharply by task complexity. Complex and contentious claims showed the lowest end-to-end AI deployment among the activities studied, while repetitive tasks showed the highest.
Provider relationships under pressure
Existing outsourcing relationships face a reset. Among BPO users, 86% said they would renegotiate or completely rework contract terms if renewals were happening today - 64% favoring renegotiation and 23% seeking a complete rework. Only 8% would renew largely as-is. Existing BPO providers ranked last among six routes insurers expect to use to acquire AI capability. Internal build and new AI managed service providers ranked highest.
Board-level scrutiny is also rising: 81% of BPO users reported a formal request or informal board or CEO-level discussion about modeling AI-driven alternatives to existing BPO contracts. AI-led innovation fell short of expectations for 49% of BPO users, the largest shortfall among nine dimensions measured, followed by strategic value at 39%. Integration remains a persistent barrier - 63% of respondents agreed that integrating AI across multiple insurance functions is a primary obstacle, while 47% ranked integration with existing technology and core systems among their top three barriers.
Why this matters for insurance professionals
The 13% figure is the number to watch. With 52% of insurers aiming for an advanced AI posture, the next two years will determine which organizations close the gap - and which ones stall. For underwriters, claims leaders and brokers, the immediate question is whether their firm has a clear, documented boundary between AI execution and human judgment, or whether that line is still implicit. The report suggests that without explicit governance rules tied to policy wording and decision authority, scaling AI will remain stuck at the incremental efficiency stage. Capacity constraints that leave one in nine submissions unquoted are a growth problem, not a technology problem. The firms that solve the governance question first will likely be the ones that convert AI ambition into underwriting capacity.
Your membership also unlocks:
